Institut de Recerca en Economia Aplicada Regional i Pública Document de Treball 2015/19 1/43 Research Institute of Applied Economics Working Paper 2015/19 1/43
“Evaluation of the Impact of Bus Rapid Transit on Air Pollution”
Germà Bel and Maximilian Holst
4
WEBSITE: www.ub.edu/irea/ • CONTACT: [email protected]
The Research Institute of Applied Economics (IREA) in Barcelona was founded in 2005, as a research
institute in applied economics. Three consolidated research groups make up the institute: AQR, RISK
and GiM, and a large number of members are involved in the Institute. IREA focuses on four priority
lines of investigation: (i) the quantitative study of regional and urban economic activity and analysis of
regional and local economic policies, (ii) study of public economic activity in markets, particularly in the
fields of empirical evaluation of privatization, the regulation and competition in the markets of public
services using state of industrial economy, (iii) risk analysis in finance and insurance, and (iv) the
development of micro and macro econometrics applied for the analysis of economic activity, particularly
for quantitative evaluation of public policies.
IREA Working Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. For that reason, IREA Working
Papers may not be reproduced or distributed without the written consent of the author. A revised version
may be available directly from the author.
Any opinions expressed here are those of the author(s) and not those of IREA. Research published in
this series may include views on policy, but the institute itself takes no institutional policy positions.
Abstract
Mexico City’s bus rapid transit (BRT) network, Metrobus, was introduced in an attempt to reduce congestion, increase city transport efficiency and cut air polluting emissions. In June 2005, the first BRT line in the metropolitan area began service. We use differences-in-differences and quantile regression techniques in undertaking the first quantitative policy impact assessment of the BRT system on air polluting emissions. The air pollutants considered are carbon monoxide (CO), nitrogen oxides (NOX), particulate matter of less than 2.5 µm (PM2.5), particulate matter of less than 10 µm (PM10), and sulfur dioxide (SO2). The ex-post analysis uses real field data from air quality monitoring stations for periods before and after BRT implementation. Results show that BRT constitutes an effective environmental policy, reducing emissions of CO, NOX, PM2.5 and PM10.
JEL classification: Q51, Q58, R41, R48 Keywords: Bus Rapid Transit, Differences-in-Differences, Environmental Policy Evaluation, Public Transport, Urban Air Pollution
Germà Bel: Department of Economic Policy & GiM-IREA, Universitat de Barcelona (Barcelona, Spain) ([email protected]). Maximilian Holst: Department of Economic Policy & GiM-IREA, Universitat de Barcelona (Barcelona, Spain) ([email protected]) Acknowledgements This work was supported by the Spanish Government under the project ECO2012-38004; the Catalan Government under project SGR2014-325, and the ICREA-Academia program of the Catalan Government.
1
Evaluation of the Impact of Bus Rapid Transit on Air Pollution I. Introduction
In the literature of environmental and transport economics, road transport is widely considered
one of the main sources of air pollution. More specifically, a large fraction of GHG emissions
and air pollutants are recognized as being derived from road traffic: “In 2004, transport
accounted for almost a quarter of carbon dioxide (CO2) emissions from global energy use.
Three-quarters of transport-related emissions are from road traffic” (Woodcock et al., 2009, p. 2).
Moreover, these pollution levels are particularly high in areas that suffer severe levels of traffic
congestion. Conventional road transport produces a series of pollutant emissions, which in high
concentrations represent a hazard for the inhabitants of urban areas. The most usual pollutants
are particulate matter of different size fractions (PM10 and PM2.5), carbon monoxide (CO), sulfur
dioxide (SO2), nitrogen oxides (NOX), and carbon dioxide (CO2). Combustion engines do not
necessarily produce all these pollutants, but some of the emissions from these engines in
combination with other particles in the air can react with more complex molecules (such as,
ozone) and have a negative impact on human health.
Road transit, as a major determinant of air pollution in urban areas, can be broken down
into different sectors, with one of the most relevant being that of public transport. Urban buses
emit relatively high levels of CO, NOX, PM10, and CO2. However, due to the use of cleaner, better
quality fuels and to stricter regulations on road traffic emissions, the net air quality impact of
buses can be positive if vehicles are replaced periodically. This is particularly true if cities adopt
electric vehicles and this energy is generated from renewable sources.
Public transport systems, such as subways and or light rail networks, are emission friendly
transport options that are able to transport huge numbers of people on a daily basis. The
downside of these modes of transportation, however, is the enormous initial investment they
require and the rigidity of their services. Most governments operate under considerable budget
2
constraints so that building or expanding local public transport infrastructure requires massive
investment, while construction is not always feasible owing to the nature of the local geography.
In the last few decades, governments have sought alternatives that are similarly effective
but at the same time more affordable. One such option is the Bus Rapid Transit (BRT) system, a
high-quality bus service with a similar performance to that of a subway, but provided at a fraction
of the construction cost (Cervero, 1998). Many countries around the world have adopted BRT
systems. The main factors in their favor are the low initial investment costs (especially compared
to a subway line), low maintenance costs, operating flexibility, and the fact that they provide a
rapid, reliable service (Deng & Nelson, 2011). If a BRT line is unable to capture the projected
transport demand, or if the usual route is under maintenance, the line can easily be rerouted.
The literature addressing the impact of BRT on air quality does not quantify the reduction
in concentrations of the different pollutants. Most assessments are qualitative studies of impact
effects or take the form of fuzzy cost-benefit analyses that fail to provide details about individual
pollutant levels. Our research seeks to address this gap in the literature. The contributions of this
paper are, as such, easily identifiable: a) to provide a rigorous quantification of the impact on air
quality of the introduction of a BRT network in a metropolitan area; b) to add to the few analyses
to date that employ actual field data in their evaluations of public transport policy; and c) to
employ econometric-based methods of differences-in-differences and quantile regression to
analyze the environmental impact of a public transportation system like BRT.
II. Related Literature
Several studies have examined the impact of pollutants and report the potential effects for health.
PM10 and PM2.5 have been linked with a decrease in respiratory capacity, aggravating asthmatic
conditions, and with severe heart and lung damage (WHO, 2001). Nitrogen oxides (NOX), and
particularly nitrogen dioxide (NO2), affect the respiratory system and intensify existing cases of
pneumonia or bronchitis, while NOX in high concentrations can seriously damage lung tissue.
Sulfur dioxide (SO2) can worsen existing symptoms of respiratory or cardiovascular diseases.
3
Carbon monoxide (CO) is one of the most common types of poisoning. It can disable the
transport of oxygen to the cells and cause dizziness, headaches and nausea; in high
concentrations can lead to unconsciousness and death (Neidell, 2004; Schlenker & Walker, 2011).
Moreover, PM10 is considered a risk factor for respiratory related post-neonatal mortality
and sudden infant death syndrome (Woodruff et al., 2008). The effects of alleviating traffic
congestion on infant health are analyzed extensively in Currie & Walker (2011), who show that a
reduction in congestion increases the health and development of infants significantly (see also
Kampa & Castanas, 2008; Wilhelm & Ritz, 2003; Wilhelm et al., 2008; and Lleras-Muney, 2010).
Many institutions are aware that substantial government efforts are needed to initiate
change and have accepted the challenge of fighting the problems of air pollution. And, indeed,
many governments have introduced policies to reduce the emissions generated by their services.
For example, in 2009, the São Paulo city council approved the Municipal Policy for Climate
Change, aimed at reducing GHG emissions by 20% in 2020, taking 2005 as its baseline (Lucon &
Goldemberg, 2010, p. 348). In this instance, the council’s measures focused on transportation,
renewable energy, energy efficiency, waste management, construction and land use.
Some governments have specifically targeted road traffic pollutants (World Resources
Institute, 2011). For example, in 2009, the Japanese central government announced a USD $154
billion package to foster environmentally friendly technologies. Among others, the package gives
incentives (tax breaks worth as much as $2,500) to automobile consumers for the purchase of
hybrid/electric cars, as well as subsidies of 5% on other energy efficient consumer goods. In
Germany, the government introduced low emission zones (LEZ) in many cities. Using
differences-in-differences, Wolff (2013) finds that the LEZs managed to reduce emission of PM10
by 9%.
An alternative policy for abating emissions from road traffic is the introduction of
maximum speed limits on highways or in certain metropolitan areas. Many studies have examined
the impact of such policies by employing a vast range of analytical techniques. The majority
4
calculate the impact on pollution rates resulting from changes at a local level. However, it is
implicitly assumed that no other factors play a role and, thus, the changes are summed at an
aggregate level. Moreover, the computations are often made ex-ante. In the literature, we find
Gonçalves et al. (2008), who report modest reductions of polluting emissions in Barcelona;
Keuken at al. (2010), who find a substantial reduction in polluting levels in the Netherlands; and,
Keller at al. (2008), who estimate a 4% reduction in NOX due to this policy in Switzerland.
An alternative way of evaluating the impact of a policy on pollution levels is to measure
the effect ex-post using field data. However, few studies of this type have been reported to date.
Exceptions include Bel & Rosell (2013) on the impact of an 80km/h speed limit and a variable
speed limit policy in the metro-area of Barcelona. They report that the variable speed policy was
much more effective, reducing NOX and PM10 emissions by 7.7–17.1% and 14.5–17.3%
respectively. Similarly, Van Benthem (2015) analyzed speed limits on the U.S. West Coast, and
concludes that the optimal speed, considering costs and benefits, is about 88km/h (55 mph) and
that increasing the speed would increase CO, NOX, and O2 levels. Note that Bel & Rosell (2013)
and Van Benthem (2015) use real field data; thus, they are able to measure the actual policy
impact rather than making computations based on a series of assumptions.
This paper contributes to the existing literature by providing a robust quantification of
the impact on air quality of the BRT network in a metropolitan area. We employ actual field data
in our evaluation, and use econometric-based methods of differences-in-differences and quantile
regression to analyze the environmental impact of the Bus Rapid Transit System in Mexico.
III. Bus Rapid Transit in Mexico City
Bus Rapid Transit and pollution
Bus Rapid Transit –BRT- is a relatively new mode of public transportation that has found broad
acceptance in developing countries since the early 1990s. By the end of 2014, 186 cities around
the world had adopted some form of BRT. We find prominent examples in Bogotá, Curitiba,
Guangzhou, Jakarta, and Istanbul. Latin America is seen as the epicenter of the global BRT
5
movement (Cervero, 2013) with over 60 cities using BRT, moving about 20 million people each
day; that is, 62% of the global demand for BRT services. Above all, cities in Brazil (34), Mexico
(9) and Colombia (6) have led the rapid growth of BRT networks in the region. BRT has also
developed in Europe and the U.S. Over 50 cities in Europe provide this service to an average of
2 million people daily. BRT systems exist in 18 cities in the US, transporting an average of almost
half a million people daily (see http://brtdata.org/) for figures and statistics on BRT cities).
A key feature of BRT is that it acts not only as a transport policy, but also forms part of a
country’s environmental policy. In this latter regard, it needs to be borne in mind that old buses
are being replaced by modern vehicles run on cleaner fuels, while the introduction of BRT lines
should also reduce congestion. According to Cervero (2013, p. 19), BRT is ‘likely’ to have net
benefits regarding emissions: “BRT generally emits less carbon dioxide than LRT [light rail train]
vehicles due to the use of cleaner fuels”. Cervero & Murakami (2010) consider that attracting
former motorists to BRT can reduce vehicle kilometers traveled and thus polluting emissions.
The reduction in emission levels thanks to the introduction of BRT systems is noticeable.
In Bogotá’s TransMilenio, Hidalgo et al. (2013) estimate health-cost savings from reduced
emissions following the completion of TransMilenio’s first two phases at US$114 million over a
20-year period, based on a rough computation of data. They calculate that about 8% of total
benefits can be attributed to air pollution and traffic accident savings (that is, reductions in
associated illnesses and deaths). However, the authors do not use real field data to quantify the
pollution-reduction benefits. Indeed, in Bogotá, the buses displaced by the BRT were reallocated
to the urban edge and smaller surrounding townships, leading Echeverry et al. (2005) to argue
that BRT may not have reduced the problem of polluting emissions but simply displaced it to
other areas.
Geography and Institutions
Mexico City is one of the most heavily populated metropolitan areas in the world. The estimated
population in 2005 was 19.2 million inhabitants, growing to over 20 million by 2010 (population
6
density was estimated at 2560 inhabitants/km²). The city has a subtropical highland climate and
occupies a valley at 2,220 meters above sea level. Diurnal temperatures oscillate between 10 and
22°C, and can easily climb above 30°C on hot days and fall to freezing on cold winter days.
Rainfall is intense from June to October, but it is scarce from November to May. Pollution levels
are much higher during the dry season. Wind speed plays a critical role in the city’s weather and
pollution levels: weak winds and the shape of the valley do not allow air pollutants to disperse.
The city hosts many different modes of public transport, including an extensive metro
network, light rail, buses, trolleybuses, micro-buses, taxis, etc. All modes are regulated by Mexico
City’s Mobility Secretary (SEMOVI, Mexico City Government). For several years, most modes of
public transport have operated at full capacity, resulting in lengthy commuting times, e.g., subway
commuters will typically have a long wait and have to let several trains pass before they can
board. Metro, buses and micro-buses are typically perceived as serving the lower socioeconomic
classes, as they are constantly overloaded, offer poor quality service, and due to an increasing
income gap. Those who can afford a car prefer to use it for their daily commute. Crôtte et al.
(2009) show that Mexico’s metro users that earn low wages and do not own a car perceive the
metro as a normal good, while middle/high income earners perceive the metro as inferior good.
Many bus lines serve Mexico City’s main streets and avenues. In certain cases, several bus
and micro-bus lines overlap, resulting in chaos and congestion because of the extremely slow
speeds attained and the constant stopping and starting of the bus units. Av. de los Insurgentes, one
of the longest avenues in the world at 28.8 km, and the city’s main north-to-south arterial route
used to be especially affected by congestion. The city’s public micro-bus lines suffer from an
absence of effective regulations, which means there are no official bus stops and drivers can stop
anywhere to let people on and off. The congestion attributable to the micro-buses exacerbated
commuting times. At peak hours, a commuter could take two hours to travel a distance of just 20
kilometers. This was the situation by the early 2000s, before the BRT operations were introduced.
7
The Metrobus Policy
The pollution problem is not new for Mexico City. Over the years the government tried to
implement programs aimed at reducing pollution levels. The most known one was the ‘Hoy no
circula’ (today you do not circulate) program introduced in 1989, according to which cars that do
not fulfill emission criteria could not circulate on one particular day during the week depending
on the last number of their license plate. Analyzing the impact of this program with a regression
discontinuity design, Davis (2008, p. 40) showed that this policy is not effective, but it also “led
to an increase in the total number of vehicles in circulation as well as a change in the composition
of vehicles toward high-emissions vehicles”.
On 5 November 2002, the governor of Mexico City announced an ambitious program to
deal with the worst cases of congestion. The aim was to reduce commuting times and to tackle
the city’s air quality problems, and several policies were implemented. In 2004 a few buses from
the public network were renewed. In 2006-07 some parts of the ‘second floor’ of the inner-city
highway Anillo Periférico were inaugurated. This helped reducing congestion in some areas, but the
overall amount of cars using both levels increased; so reduction of emissions was not significant.
Other minor policies were introduced in 2007, such as a pilot project of a bicycle program. All in
all, results obtained with these different programs and measures were modest.
At the heart of the 2002 program lay the introduction of a BRT (‘Metrobus’) system,
designed to reduce traffic and air pollutant emissions. The intention was not to compete with
existing public modes of transport; rather, BRT was seen as an alternative to existing options in
order to reduce congestion. Note that, as found by Anderson (2014) for Los Angeles, congestion
relief benefits alone may justify transit infrastructure investments. On March 2005, SEMOVI
oversaw the creation of the public entity Metrobus, with an initial operating budget of MXN 42.4
million pesos (USD 3.8 M in 2005). Metrobus was to be fully responsible for the BRT’s operation
planning and its control and administration.
8
The main idea underpinning the BRT system was to create an exclusive bus lane in which
only authorized buses could operate subject to certain rules and criteria (schedule time,
designated stops, physical dimensions of buses, and amount of emissions), to guarantee efficient
operation. To promote the system, several stations had to be built to enable passengers to access
the service. The project was implemented in 2005 with an initial investment of around USD $80
million to build up the infrastructure (Schipper et al., 2009). The investment included the
construction of 37 BRT stations and exclusive bus lanes and the introduction of new articulated
buses run on conventional diesel fuel. BRT was first opened on Av. de los Insurgentes; the first line
in this corridor was 19.6 km long (it was extended to 28.1 km in 2008). BRT lanes reduced traffic
congestion, as the measure eliminated overlapping of services with other bus lines. At the same
time, flow in the car lanes was improved as traffic no longer had to stop whenever a bus stopped.
Following the introduction of the Metrobus, the city’s old buses and micro-buses
operating on the same BRT route were reallocated or simply scrapped. The substitution of these
old units represented an important change in terms of the air quality conditions in the areas
adjacent to the new Metrobus route. Micro-buses, often allowed to operate because of the
authority’s negligence, represented one of the main sources of health-threatening gases for the
population. The aim of the policy was to lower the air polluting emissions of public
transportation, and the units operating the BRT network satisfy specific standards (Euro V
emission standard).
The analysis of historical trends of energy demand, air pollutants and GHG emissions
attributable to passenger vehicles commuting in Mexico City’s metro-area done by Chavez-Baeza
& Sheinbaum-Pardo (2014), reported that the primary sources of small particle matter are road
passenger transport vehicles. According to in-vehicle measurements by Shiohara et al. (2005),
carcinogenic risks caused by micro-buses were much higher than those caused by buses and the
metro. In a related study, Gómez-Perales et al. (2004) measured (in-vehicle) commuters’ exposure
to PM2.5, CO and benzene in micro-buses, buses and the metro in Mexico City during morning
9
and evening rush hours. They reported that pollution levels inside the micro-bus units presented
the highest concentrations for all the pollutants during rush hours. Wöhrnschimmel et al. (2008)
compared micro-bus, regular bus and BRT unit emissions in Mexico City. Based on in-vehicle
emission measurements, they concluded that Metrobus units were the least polluting of the three
options given that the buses are newer, more efficient and run on diesel instead of regular fuel.
While it seems intuitive that there is less pollution because of vehicle substitution, it is not
clear whether pollution levels in the metropolitan area have also been reduced. Less congestion
on a particular route may induce more people to use it. Hence, an increase in demand may even
increase pollution levels in a given area if a sufficient number of commuters are attracted to use
it. According to the Metrobus office, standard commuting times have fallen from 1 hour 30
minutes to 1 hour on the route, while passenger exposure to benzene, CO, and PM2.5 has fallen
by up to 50 percent, compared to the figures for the previous bus service operating in this
corridor. The office also claims that CO2 emissions have been cut by 35,000 tons per annum.
However, the accuracy of this information is questionable as these outcomes are likely to be
based on computations from in-vehicle emission changes, rather than real field data.
The Mexico City government monitors the air quality within its metropolitan area, by
measuring levels of various pollutants within its network of automatic air quality monitoring
stations distributed across the city. These stations have been operational during a number of
years and the information is made publicly available. We use this information to measure the
impact of the introduction of the Metrobus system on the concentrations of five pollutants.
(Insert Figure 1 around here)
The number of passengers using BRT has increased over the years, reaching satiation
point in some parts (see Table 1). Some years after the first line was opened, the network was
expanded, with lines two (20 km) and three (17 km) opening on December 2008 and February
2011, respectively. Line four (14 km) started operations on April 2012 and line five (10 km) on
November 2013. Metrobus network transported a total of 254 million passengers in 2014. The
10
Institute for Transportation and Development Policies (ITDP) evaluates BRT networks around
the world. In 2013 Mexico Metrobus was given a ‘Silver’ ranking according to the BRT Standard
Score indicating that it “includes most of the elements of international best practice and is likely
to be cost-effective on any corridor with sufficient demand to justify BRT investment. These
systems achieve high operational performance and quality of service” (BRT Standard 2014, p.10).
(Insert Table 1 around here)
IV. Data and variables
Pollution levels vary depending on a range of meteorological factors that have to be taken into
consideration to capture this variation. Air contaminants are not static and so the average daily
wind speed and average daily wind direction are included in the model. Wind direction is an
important factor as a significant amount of pollution might be created in heavily industrial areas
and then transported to other parts of the metropolitan area. Not only are pollutants transported,
they also undergo a number of reaction processes. The rates of these reactions are influenced by
temperature, so the average daily temperature needs to be considered. Water can result in a
reactive change in the equilibrium or it may increase sedimentation; thus, relative humidity and
daily rainfall are both included. Rainfall also reduces significantly the amount of pollutants in the
air and so this meteorological variable has to be included. Note, however, that owing to data
limitations, rainfall is calculated as the sum of daily rainfall amounts.
Data on air-related control variables (relative humidity, temperature, wind direction and
wind speed) were obtained from Mexico City’s Environment Secretary, which serves as the
official monitoring entity. Data on air quality and amount of polluting emissions come from the
Atmosphere Monitoring System (SIMAT), which comprises a network of around 40 monitoring
points distributed across the Mexico City metro-area. The SIMAT network is divided into four
monitoring subsystems, each measuring different atmospheric components and factors.
For the analysis of air pollutants, the RAMA (Automatic Network for Atmospheric
Monitoring) subsystem serves as the source for all pollutant measurements. The RAMA network
11
comprises 29 monitoring stations. The pollutants monitored are carbon monoxide (CO),
nitrogen oxides (NOX), sulfur dioxide (SO2), particles of the order of 10 micrometers or less in
aerodynamic diameter (PM10), and particles of the order of 2.5 micrometers or less (PM2.5).
Data on the meteorological parameters are obtained from the Meteorology and Solar
Radiation Network subsystem (REDMET), which comprises 19 continuous monitoring stations
that measure wind direction, wind speed, temperature, humidity, atmospheric pressure and solar
radiation. Unfortunately, data on atmospheric pressure and solar radiation are not available after
2003, which is a limitation of the model presented below.
Further data on rainfall were provided by Mexico City’s Water Systems office (SACM).
This network of rainfall measuring stations comprises 78 monitoring stations distributed across
the metropolitan area. Information on the exact location of the measuring stations was denied for
reasons of “national security”, given that details regarding the city’s waterworks infrastructure are
restricted access only. However, the names of the stations were provided and as these typically
include a reference to their location, it was possible with Google Maps to approximate the
location of most of them. Of the stations, 70.5% were easy to locate, 16.7% were roughly
approximated and 12.8% of the stations were impossible to locate based on their name.
(Insert Table 2 around here)
As the air quality monitoring stations and rainfall measuring stations did not coincide, a
matching was undertaken. Using the location of the air quality monitoring stations the closest
rainfall station within a range of less than 10 km was selected. We assume that the weather
conditions present at the air quality stations and at their closest respective rainfall stations do not
differ. The rainfall stations that could not be located are not considered here given the
impossibility of matching them to the air quality monitoring stations (the result of the station
matching is available upon request).
Our analysis of Metrobus focuses solely on line 1 (opened on 19 June 2005). We measure
its impact for the two-year period prior to its opening and the two-year post-operational period.
12
(Insert Table 3 around here)
V. Differences-in-Differences
The first part of the analysis employs the differences-in-differences method to facilitate the
measurement of the impact of the new BRT system on polluting emissions. By so doing, the
intention is to estimate the atmospheric concentration of pollutants in Mexico City between 2003
and 2007 and to assess the impact of the introduction of the Metrobus.
Methodology
The panel data used for this analysis are unbalanced. This characteristic of our panel comes from
the fact that some stations were in operation from the beginning of the period of analysis, while
other new ones were introduced at a later point in time, sometimes substituting older ones. On
the other hand, most stations required maintenance at some point. The introduction or
switching-off of the stations is exogenous and not correlated with the variables in the model.
In the absence of a randomized trial, the method adopted here is an extension of the
differences-in-differences estimation procedure specified as a two-way fixed effects model. As
stated in Wooldridg (2010: p. 828), “the usual fixed effects estimator on the unbalanced panel is
consistent”
Yit = βXit + γZit + θi + δt + εit (1)
where Yit is air pollutant concentration, Xit is a vector of time-varying control covariates that
include atmospheric characteristics, and Zit is the BRT impact dummy variable to be evaluated.
As usual in this kind of models, θi are station-specific fixed effects, δt are time-specific fixed
effects and εit is the random error. Station fixed effects control for time-invariant station-specific
omitted variables; time fixed effects control for trends around each monitoring station.
The key variable in this differences-in-differences approach is γ, which measures the
difference between the average change in air pollutant concentrations for the treatment group
(stations close to the Metrobus line) and average change in concentrations for the control group
(stations located some distance from the area through which the Metrobus passes). Specifically,
13
γ = [E(YB| BRT=1) - E(YA| BRT=1)] - [E(YB| BRT=0) - E(YA| BRT=0)] (2)
where YB and YA denote the air pollutant concentrations before and after Metrobus came into
operation. BRT=1 and BRT=0 denote treatment and control group observations respectively.
The equation for the dependent variables (CO, NOX, PM2.5, PM2.5 and SO2) is:
Yit = β0 + β1 Metrobusit + β2 Pollutant Lagit + β3 Humidityit + β4 Temperatureit + β5 Wind Directionit + β6
Wind Speedit + β7 Rainfallit + β8 Workdayt + β9 Montht + θi + δt + εit (3)
A basic assumption when using differences-in-differences is that the temporal trend in
the two areas is the same in the absence of the intervention. If this were not the case, the impact
being measured would be biased. The problem of endogeneity can also bias an impact evaluation.
According to Bertrand et al. (2004), most problems related to endogeneity can be avoided by
using the differences-in-differences technique. When using differences-in-differences in a panel
data setting, regressions must be undertaken with fixed effects: the correlation between the error
components of station i and the explanatory variables should be different from zero. Closely
related to this, an important assumption here is that unobservable variables and unobservable
characteristics remain constant over time.
In conducting the analysis the parallel trend assumption is tested to see if the parallel
trend is satisfied in the time period before treatment (i.e. before policy implementation). For the
test, the data were grouped by trimester. The mean value of each pollutant in the treated group
(within a 2.5-km radius of the Metrobus line) was then compared with the corresponding value in
the control groups. The null hypothesis is that in the absence of intervention, the trend presented
by the treated group is equal to that presented by the control group. The null hypothesis is
accepted at the 95% confidence level, indicating that the parallel trend is satisfied for all
pollutants except for PM10. Moreover, the evolution in the pollutant levels over time is provided
in graph form in Figure A1 in the Appendix. These graphs show how the treated and the non-
treated pollutant levels behaved similarly during the pre-treatment period.
14
The failure to satisfy the parallel trend assumption in the case of PM10 leads to a biased
impact evaluation for this particular pollutant. However, despite this slightly upward bias, the
PM10 analysis is included because of the importance of this pollutant. The impact evaluation of
the remaining pollutants is not biased since the parallel trends assumption is satisfied.
As mentioned, an unbalanced panel data setting requires the use of a panel fixed effects
estimation. To confirm the correct use of fixed effects in this panel, the Hausman test was run
with every pollutant. In all cases the null hypothesis of the Hausman test was rejected at the 99%
confidence level, which confirms the correct use of the method. We test the model’s basic
assumptions (homoscedasticity, time dependence, spatial dependence and exogeneity of
explanatory variables). Autocorrelation is a persistent problem for all pollutants. To account for
this problem, we included a one-period lag of the respective pollutant in each regression.
By using Driscoll-Kraay standard errors, the estimator is modified in such a way that it is
robust to cross-section and time dependence. In this way, standard errors are also
heteroscedasticity-consistent (Driscoll & Kraay, 1998). In addition, panel-corrected standard
errors (PCSE) are used to provide a robustness analysis of the results, as PCSE yield more
accurate standard errors than estimations using feasible generalized least squares (Beck, 2001).
Results
Tables 4-8 present the results for the fixed effects regression. The models for CO, NOX, PM2.5,
PM2.5 and SO2 are all jointly statistically significant at the 1% level. All estimations include year
dummies, which capture time fixed effects (coefficients for year dummies and the constant term
are not included in the outputs, and are available upon request). R² values range between 0.59-
0.61 for CO, 0.54-0.61 for COX, 0.57-0.63 for PM2.5, 0.52-0.58 for PM10, and 0.29-0.38 for SO2.
Table 4 presents the output for the fixed effects estimation of carbon monoxide. The
estimation shows a downward trend in the relationship between the impact of the introduction of
the Metrobus on pollution and distance from the Metrobus route. In areas near the BRT line, the
reduction in concentration was 19.4%, while in the areas lying between 2.5 and 10 km and
15
between 10 and 30 km from the route the reduction was 17.2% and 16.6%, respectively. The
results also identify the influence of the time lag on current levels of carbon monoxide, i.e.,
yesterday’s pollution levels determine to a large extent today’s pollution levels. A further factor
playing a key role in the levels of CO in the air is the day of the week. Thus, pollutant levels are
much higher during the week, when workers have to commute, than on the weekends.
Environmental factors such as wind and humidity also play a marked role in air pollutant
concentrations over the city as both variables are significant.
(Insert Table 4 around here)
The estimations of NOX present the opposite pattern to that presented by CO. Although
the outcome is not significant in areas close to the Metrobus route, the reduction in NOX
concentrations is greater in more distant areas. The coefficient sign is negative, which is
consistent with that of the other pollutants, and presents values between 12.2 and 18.1%. The
temporal lag plays an important role in the case of NOX, as well as in all three areas defined
around the Metrobus route. Higher wind speeds have a significant effect on the concentration
levels, blowing the pollutant into other areas when the wind speed is high. Week days have a
similar effect on pollutant concentrations as that described above for CO. For this pollutant the
year dummies are significant, capturing unobserved characteristics related to the time trend.
(Insert Table 5 around here)
Table 6 presents the output of PM2.5. In this case, only seven air quality stations monitor
this pollutant within the three areas around the Metrobus route and they are not evenly
distributed. Thus, there is only one station within a 2.5-km radius of the Metrobus route, five in
the area lying between 2.5 and 10 km from the BRT line and another one in the last zone. Due to
the small number of stations, the PM2.5 regressions in the areas with just one station are estimated
with OLS using robust standard errors. Fixed effect estimations are not feasible for these areas
since the panel structure no longer holds.
16
Bearing this in mind, distance from the route has a similar impact on concentrations to
that reported above for NOX. The introduction of the Metrobus was highly significant (at the 1%
level) in bringing the concentration levels of PM2.5 down by 20.8% in the area closest to the
Metrobus line and by 39.0% in the area lying at a distance of between 10 -30 km. The temporal
lag once again is highly significant (1%), which means that the levels of concentration of this
pollutant are also largely determined by the levels the day before. The environmental factors
affecting the concentration of PM2.5 are similar to those for NOX, with the difference that
temperature plays a more important role in the area lying up to 10 km from the route. The 5%
significance of the dummy controlling for the month of the year indicates that this variable
captures significant seasonal variations within the year.
(Insert Table 6 around here)
As noted, the results for PM10 present a slight upward bias and should be treated with
caution. However, the reduction in concentrations was substantial. In the area in the 2.5-km
radius of the Metrobus route, the PM10 level fell by 12.9 µg/m³ or 24.4% following the opening
of the line. The areas lying between 2.5 and 10 and between 10 and 30 km from the route had a
reduction of 17.7 and 15.5% in the levels of PM10, respectively (all reductions are statistically
significant). Table 7 shows how the impact on this pollutant fell with increasing distance from the
Metrobus; the reverse of the pattern presented by PM2.5 and NOX but the same as that of CO.
(Insert Table 7 around here)
Humidity levels, wind speed and week days have an influence on PM10 concentration
levels, all three being statistically significant. Higher humidity levels reduce PM10 concentrations in
the air. Week days present higher levels of pollutant concentrations than those recorded on
weekends. In the areas lying furthest from the Metrobus route (2.5-10 km and 10-30 km), the
year dummies are significant. The temporal lag of the endogenous variable indicates that past
emission levels significantly affect today’s concentration levels. The impact of the weekday
dummy is in line with the effect on the other pollutants. Commuting to work or school at peak
17
times during the week creates congestion within the city, which increases pollution levels in areas
closest to these congested roads.
Finally, our estimations of the SO2 concentrations do not show any significant effect of
the introduction of the Metrobus in any of the three areas defined around Av. de los Insurgentes. As
expected the signs of the coefficients are negative, but the variation of the error term is too high
to capture any significant impact from the Metrobus operation. Interestingly, the model for this
pollutant performs worse in terms of explanatory power, as the coefficient of determination R² is
below that of the other pollutants. It seems probable that the model is omitting other important
determinants. As above, however, the lagged value of the endogenous variable, wind and week
day variables have a significant influence on the concentration level of SO. Higher wind speeds
reduce levels of concentration while the levels rise on days when commuters take to the roads.
The estimation outputs of the different pollutant molecules show that the introduction of
the Metrobus had a marked impact on the concentration levels of the different pollutants in the
three areas defined. To appreciate better the impact of the Metrobus operation on air quality in
the Mexico City metropolitan area, Table 9 summarizes this impact for all pollutants. In the case
of NOX and PM10 the pollutant concentration increases with distance, while in the case of CO,
PM2.5 and SO2 the concentration reduces with distance. This difference might be related to the
molecular composition of each pollutant, its molecular weight, the interaction of each pollutant
with the other molecules floating in the air, and the extent to which each pollutant is affected by
environmental factors such as wind and humidity levels.
(Insert Table 8 around here)
Since estimations were carried out using Driscoll-Kraay standard errors, we also include
the results using PCSE to ensure a more robust methodological analysis. Table 9 presents the
results for the policy variable with both corrections. The policy is effective across pollutants, but
for SO2 effectiveness depends on the computation of the standard errors.
(Insert Table 9 around here)
18
VI. Quantile Regression
The differences-in-differences method using fixed effects, in common with most econometric
methods, deals with the averages of distributions. This means that what is happening in different
segments of a distribution is often ignored. In order to further our analysis of the changes in air
quality following the opening of the BRT line, we divide the sample into quantiles (more
precisely, into deciles). By so doing, the analysis becomes much more detailed and we are able to
determine which deciles of the pollutants are affected most by the introduction of the Metrobus.
Quantile regression allows us to identify whether the impact concentrates around the median or,
alternatively, at the extremes of the distribution.
Methodology
The equation specified for the quantile regression resembles that specified above for the fixed
effects regression using differences-in-differences:
Q Yit (τ) = β(τ)Xit + φ(τ)Zit + θi + δt (4)
where Q Yit (τ) is the quantile function at confidence level τ. This model allows the influence of
the control variables Xit and the policy variable Zit to depend on the quantile confidence level τ.
Again, θi and δt are station-specific and time-specific fixed effects. To estimate this model,
Koenker (2004) proposes the simultaneous estimation of the following equation:
min(β, γ, θ) ∑(q=1…Q)∑ (i=1…n)∑ (t=1…T) wq ρτq (Yit - β(τ)Xit - φ(τ)Zit - θi - δt) (5)
where ρτq(·) is the function below (as in Koenker & Bassett, 1978; see also Koenker, 1984):
ρτ(u)= τ|u|, u ≥ 0 (6) (1 - τ)|u|, u < 0
The term wq are chosen weights and they control the influence of the quantiles on the
estimation of the fixed effects. Note that neither the Gaussian condition nor the classical
hypothesis related to the random error term is necessary here. Bel et al. (2015), who suggest this
way of proceeding, stress this aspect about the error term. In common with these authors, we
also assume that the weights are the same for all the quantiles analyzed. As discussed above, the
19
quantile regression expression can be seen as the differences-in-differences model decomposed
into quantiles (deciles). Therefore for any given confidence level τ,
φ = [Q(YB| BRT=1) - Q(YA| BRT=1)] - [Q(YB| BRT=0) - Q(YA| BRT=0)] (7)
where YB and YA denote the air pollutant concentrations before and after the introduction of
Metrobus. As in the first analysis, here we seek to estimate the differences between the treated air
quality monitoring stations and the stations that lie furthest away from the BRT system (control
group), while considering the changes in emissions before and after introducing Metrobus.
Recall that for the quantile regressions robust standard errors have also been used. The
robust standard errors are computed under the assumption that the residual density is continuous
and bounded away from 0 and infinity at the specified quantile (Koenker, 2005).
Results
Tables 10-12 show the results of the quantile regressions. The results of the diff-in-diff analyses
suggested that some pollutant impacts were not always significant in the three areas defined
around the Metrobus corridor. Now we determine if the areas that did not register any significant
impact on a pollutant did in fact experience some effect in some parts of the distribution.
Table 10 presents the results for the area lying closest to the Metrobus route. We obtain a
negative sign across all pollutants and deciles of the distribution. Interestingly, CO, NOX and
PM10 levels are significantly affected across the distribution, while PM2.5 shows a significant
impact in the lower deciles, becoming weaker at the upper end. SO2, which did not present
significant outcomes when using the Driscoll-Kraay standard errors, presents a significant impact
in this area at the upper end of the distribution, with the concentration level down by 63.9%.
(Insert Table 10 around here)
The results for the area lying between 2.5 and 10.0 km from the line (Table 11) are
consistent with those from the differences-in-differences analysis for CO, NOX and PM2.5. In this
area, PM10 presents significant values around the median of the distribution but not at the bottom
end. This pollutant presented significant outcomes in the differences-in-differences analysis, but
20
here we see that the impact is located around the middle and upper parts of the distribution. In
common with the area adjacent to the BRT line, this area presents negative and highly significant
results for SO2. Again, when using Driscoll-Kraay standard errors this pollutant did not present
any significant results and so these results would be more in line with the PCSE estimation.
(Insert Table 11 around here)
Table 12 shows the results for the area lying at a distance of between 10-30 km from the
Metrobus. This area presents very similar results to those in the second zone (2.5 to 10 km). CO,
NOX and PM2.5 are significant in all parts of the distribution while PM10 shows significant results
around the median. SO2, on the other hand, is significant around the median and at the lower
part of the distribution.
(Insert Table 12 around here)
VII. Conclusions
This paper has evaluated the impact of the introduction of Bus Rapid Transit on pollution levels
in Mexico City. The analysis has been based on real field data obtained from automatic air quality
monitoring stations and has focused on five pollutants: CO, NOX, PM2.5, PM10 and SO2.
Using unbalanced panel data, we conduct an impact evaluation using the econometric-
based techniques of differences-in-differences and quantile regression, an approach not
previously used to quantify the environmental impact of this mode of transport. Results from the
differences-in-differences analysis show a significant reduction in the concentrations of all
pollutants, but SO2. Specifically, CO concentrations were reduced by 16.6-20.4%, NOX by 12.9-
18.1%, PM2.5 by 20.8-39.0% and PM10 by 9.6-24.4%, according to the city area.
In the case of SO2, the results are inconclusive. The estimation based on Driscoll-Kraay
standard errors failed to reveal any significant impact of the introduction of BRT; however, the
estimation using panel-corrected standard errors showed a significant reduction of 27.7% within
a 2.5-km radius of the Metrobus and a reduction of 23.1% in the area lying between 2.5 and 10
km from the BRT corridor.
21
The quantile regressions conducted identify the levels of the distribution at which the
policy had most impact. In the area within a 2.5-km radius of the Metrobus, the results for CO,
NOX and PM10 are significant for almost all selected quantiles of the distribution, while PM2.5 is
significant only in the lower half of the distribution (recall PM2.5 was highly significant according
to the differences-in-differences test). It is interesting to note that SO2, for which the differences-
in-differences estimation using Driscoll-Kraay standard errors showed no significance and the
estimation using PCSE revealed an impact, is significant only in the upper levels of the
distribution. However, the significance at the upper extreme is not sufficient to make the
differences-in-differences analysis with Driscoll-Kraay standard errors significant. These results
are, nevertheless, in line though with those provided by PCSE analysis.
It would be inappropriate to generalize the impact of BRT on air quality reported here to
all cities. Geographical and atmospheric traits obviously differ from one location to another.
Future research would benefit from comparing the reduction in emissions reported here with
those detected in other metropolitan areas based on real field data, and from determining
whether the latter are consistent with the findings herein. Similarly, future studies might build on
the present model and include additional environmental factors such as atmospheric pressure and
congestion monitoring variables.
For cities with similar characteristics to those of Mexico City, our results should
encourage the expansion of their BRT networks, the continuous introduction of cleaner BRT-
units, and an increase in the size of their BRT fleets to provide a better standard of service,
measures that should motivate more people to switch from private cars to public transport. It is
important to recall, however, that the emission impact of each BRT line will be different for
every corridor, and that other factors are likely to play a role. In short, decision makers that are
truly committed to the climate change fight should consider BRT as a public transport option,
and analyze whether it meets their city’s needs.
22
References
Anderson, M., 2014. Subways, Strikes, and Slowdowns: The Impacts of Public Transit on Traffic
Congestion, American Economic Review, 104, 2763–2796.
Beck, N., 2001. Time-series-cross-section data: what have we learned in the past few years?
Annual Review of Political Science, 4, 271–293.
Bel, G., Bolancé, C., Guillén, M. & Rosell, J., 2015. The environmental effects of changing speed
limits: A quantile regression approach, Transportation Research Part D, 36, 76–85.
Bel, G. & Rosell, J., 2013. Effects of the 80 km/h and variable speed limits on air pollution in the
metropolitan area of Barcelona, Transportation Research Part D, 23, 90–97.
Bertrand, M., Duflo, E. & Mullainathan, S., 2004. How much should we trust differences-in-
differences estimates? The Quarterly Journal of Economics, 119, 249-275.
Cervero, R., 1998. The transit metropolis: a global inquiry. Washington, D.C.: Island Press.
Cervero, R., 2013. Bus Rapid Transit (BRT). An efficient and competitive mode of public
transport. 20th ACEA Scientific Advisory Group Report.
Cervero, R. & Murakami, J., 2010. Effects of built Environments on vehicle miles traveled:
evidence from 370 US metropolitan areas, Environment and Planning A, 42, 400-418.
Chávez-Baeza, C. & Sheinbaum-Pardo, C., 2014. Sustainable passenger road transport scenarios
to reduce fuel consumption, air pollutants and GHG (greenhouse gas) emissions in the
Mexico City Metropolitan Area. Energy, 66, 624–634.
Crôtte, A., Noland, R. & Graham, D., 2009. Is the Mexico City metro an inferior good?, Transport
Policy, 16, 40–45.
Currie, J. & Walker, R., 2011. Traffic Congestion and Infant Health: Evidence from E-ZPass,
American Economic Journal: Applied Economics, 3, 65-90.
Davis, L., 2008. The Effect of Driving Restrictions on Air Quality in Mexico City, Journal of
Political Economy, 116, 38-81.
23
Deng, T. & Nelson, J., 2011. Recent developments in bus rapid transit: a review of the literature.
Transport Reviews, 31, 69-96.
Driscoll, J. & Kraay, A., 1998. Consistent Covariance Matrix Estimation with Spatially
Dependent Panel Data, The Review of Economics and Statistics, 80, 549-560.
Echeverry, J., Ibáñez, A., Moya, A. & Hillón, L., 2005. The Economics of Transmilenio, A Mass
Transit System for Bogotá, Economia, 5, 151–96.
Gómez-Perales, J., Colvile, R., Nieuwenhuijsen, M., Fernández-Bremauntz, A., Gutiérrez-
Avedoy, V., Páramo-Figueroa, V., Blanco-Jiménez, S., Bueno-López, E., Mandujano, F.,
Bernabé-Cabanillas, R. & Ortiz-Segovia, E., 2004. Commuters’ exposure to PM2.5, CO, and
benzene in public transport in the metropolitan area of Mexico City, Atmospheric Environment,
38, 1219–1229.
Gonçalves, M., Jiménez-Guerrero, P., López, E. & Baldasano, J., 2008. Air quality models
sensitivity to on-road traffic speed representation: effects on air quality of 80 km/h speed limit
in the Barcelona Metropolitan area, Atmospheric Environment, 42, 8389–8402.
Hidalgo, D., Pereira, L., Estupiñán, N., & Jiménez, P., 2013. TransMilenio BRT system in
Bogotá: high-performance and positive impacts; main results of an ex-post evaluation. Research
in Transportation Economics, 39, 133-138.
INEGI, 2010. Censo de población y vivienda 2010 – Estados Unidos Mexicanos Resultados Preliminares,
Instituto Nacional de Estadística y Geografía, ed.
Kampa, M. & Castanas, E., 2008. Human health effects of air pollution, Environmental Pollution,
151, 362-367.
Keller, J., Andreani-Aksoyoglu, S., Tinguely, M., Flemming, J., Heldstab, J., Keller, M., Zbinden,
R. & Prevot, A., 2008. The impact of reducing the maximum speed limit on motorways in
Switzerland to 80 km/h on emissions and peak ozone, Environmental Modelling & Software, 23,
322–332.
24
Keuken, M.P., Jonkers, S., Wilmink, I.R. & Wesseling, J., 2010. Reduced NOx and PM10
emissions on urban motorways in The Netherlands by 80 km/h speed management, Science of
the Total Environment, 408, 2517–2526.
Koenker, R., 2004. Quantile regression for longitudinal data, Journal of Multivariate Analysis, 91,
74–89.
Koenker, R., 2005. Quantile Regression, Cambridge: Cambridge University Press.
Koenker, R. & Bassett, G., 1978. Regression quantiles, Econometrica, 46, 33–50.
Lleras-Muney, A., 2010. The needs of the army: using compulsory relocation in the military to
estimate the effect of air pollutants on children's health, Journal of Human Resources, 45, 549–
590.
Lucon, O. & Goldemberg, J., 2010. São Paulo—The “Other” Brazil: Different Pathways on
Climate Change for State and Federal Governments, The Journal of Environment Development, 19,
335-357
Neidell, M., 2004. Air pollution, health, and socio-economic status: the effect of outdoor air
quality on childhood asthma, Journal of Health Economics, 23, 1209–1236.
Schlenker, W. & Walker, W., 2011. Airports, air pollution, and contemporaneous health, NBER
Working Paper, No. 17684.
Shiohara, N., Fernández-Bremauntz, A., Blanco-Jiménez, S. & Yanagisawa, Y., 2005. The
commuters’ exposure to volatile chemicals and carcinogenic risk in Mexico City, Atmospheric
Environment, 39, 3481–3489.
Van Benthem, A., 2015. What Is the Optimal Speed Limit on Freeways? Journal of Public Economics,
124, 44-62.
WHO, 2001. Quantification of health effects of exposure to air pollution. Report on a WHO Working Group.
Bilthoven, Netherlands, 20–22 November 2000. WHO Regional Office for Europe,
EUR/01/5026342 (E74256).
25
Wilhelm, M., Meng, Y., Rull, R., English, P., Balmes, J. & Ritz, B., 2008. Environmental public
health tracking of childhood asthma using California Health Interview Survey, Traffic, and
Outdoor Air Pollution Data, Environmental Health Perspectives, 116, 1254–1260.
Wilhelm, M. & Ritz, B., 2003. Residential proximity to traffic and adverse birth outcomes in Los
Angeles County, California, 1994–1996. Environmental Health Perspectives, 111, 207–216.
Wöhrnschimmel, H., Zuk, M., Martínez-Villa, G., Cerón, J., Cárdenas, B., Rojas-Bracho, L. &
Fernández-Bremauntz, A., 2007. The impact of a Bus Rapid Transit system on commuters’
exposure to Benzene, CO, PM2.5 and PM10 in Mexico City, Atmospheric Environment, 42,
8194–8203.
Wolff, H., 2013. Keep your clunker in the suburb: low-emission zones and adoption of green
vehicles, The Economic Journal, 124, F481–F512.
Woodcock, J., Edwards, P., Tonne, C., Armstrong, B., Ashiru, O., Banister, D., Beevers, S.,
Chalabi, Z., Chowdhury, Z., Cohen, A., Franco, O., Haines, A., Hickman, R., Lindsay, G.,
Mittal, I., Mohan, D., Tiwari, G., Woodward, A., Roberts, I., 2009. Health and Climate
Change 2: Public health benefits of strategies to reduce greenhouse-gas emissions: urban land
transport, The Lancet, 374, 1930-1943.
Woodruff, T., Darrow, L. & Parker, J., 2008. Air pollution and post neonatal infant mortality in
the United States, 1999–2002, Environmental Health Perspectives, 116, 110–115.
Wooldridge, J., 2010. Econometric Analysis of Cross Section and Panel Data, Massachusetts: MIT Press.
World Resources Institute, 2011. A Compilation of Green Economy Policies, Programs, and
Initiatives from Around the World. The Green Economy in Practice: Interactive Workshop 1,
February 11th, 2011.
26
FIGURES
Figure 1: Metrobus Line 1 and the air quality monitoring stations in Mexico City’s metro-area
27
TABLES
Table 1: Number of passengers using Mexico-City’s Metrobus Network
Year Line 1 Line 2 Line 3 Line 4 Line 5 Total
2005 31,515,511 0 0 0 0 34,720,301
2006 74,321,914 0 0 0 0 74,218,369
2007 77,505,395 0 0 0 0 77,652,053
2008 89,201,679 1,891,080 0 0 0 89,804,339
2009 93,455,128 33,869,530 0 0 0 127,134,909
2010 99,342,235 38,187,092 0 0 0 136,915,678
2011 113,046,246 43,469,130 32,954,167 0 0 187,183,000
2012 122,082,471 47,364,386 39,890,301 10,982,706 0 220,319,864
2013 124,891,960 48,078,130 40,546,259 13,599,680 3,157,914 230,273,943
2014 124,560,033 47,995,096 42,072,979 18,171,539 21,209,779 254,009,426
Source: Data from the Metrobus Public Information Office
28
Table 2: Description and Source of the model variables
Variable Description Source
CO Carbon Monoxide daily average concentration (ppm) RAMA
NOX Nitrogen oxides daily average concentration (ppm) RAMA
PM2.5 Particulate Matter with less than 2.5 µm (µg/m³) daily
average concentration RAMA
PM10 Particulate Matter with less than 10 µm (µg/m³) daily
average concentration RAMA
SO2 Sulfur Dioxide daily average concentration (ppm) RAMA
CO(-1), NOX(-1), PM2.5(-1),
PM10(-1), SO2(-1) One period lag (1 day) of the polluting variables RAMA
Metrobus Binary variable: 1 if the Metrobus is implemented, 0
otherwise.
Metrobus Public
Information Office
Relative humidity Daily average relative humidity (%) REDMET
Temperature Daily average temperature (°C) REDMET
Wind Direction Daily average wind direction (Azimuth Degrees) REDMET
Wind speed Daily average wind speed (m/s) REDMET
Rainfall Sum of the daily rainfall (mm) SACM
Weekdays Binary variable: 1 if the day is a labor day (Monday-
Friday), 0 if day is a Saturday or a Sunday.
Note: ppm = parts per million; µg/m³ = micrograms per cubic meter; m/s = meters per second; mm =
millimeters
29
Table 3: Descriptive statistics of the model variables
Variable Mean Std. Deviation Min. Max. Obs. Stations
CO 1.294 0.601 0.39 6.84 23.589 17
NOX 59.444 30.011 3.75 241.65 24.139 17
PM2.5 27.515 12.629 5.22 160.75 9.528 7
PM10 51.397 25.074 1.67 318.29 17.925 14
SO2 9.928 9.928 0.86 115 29.935 23
Metrobus 0.5 0.5 0 1 1.461 -
Relative humidity 56.461 12.44 24.74 87.23 16.491 18
Temperature 16.194 2.406 7.45 23.57 15.469 18
Wind Direction 186.96 23.53 116.4 295.93 16.612 17
Wind speed 1.74 0.449 0.92 3.84 16.612 17
Rainfall 1.633 2.877 0 18.88 113.958 78
Weekdays 0.714 0.452 0 1 1461 -
30
Table 4: Estimation of the logarithm of Carbon Monoxide (CO) daily average concentration
Dependent Variable: (1) (2) (3)
Log(CO) 0 - 2.5 km 2.5 - 10.0 km 10.0 - 30.0 km
Metrobus -0.1940 ** -0.1720 *** -0.1660 **
(0.0506) (0.0394) (0.0456)
Temporal lag: Log(CO) 0.5780 ** 0.5340 *** 0.5570 ***
(0.0353) (0.0217) (0.0223)
Humidity 0.0049 * 0.00381 ** 0.0060 **
(0.00206) (0.00157) (0.00193)
Temperature 0.0045 -0.0108 0.0011
(0.0123) (0.00792) (0.00966)
Wind Direction -0.0004 0.0000954 -0.0007 *
(0.000234) (0.000168) (0.000278)
Log(Wind Speed) -0.4060 *** -0.444 *** -0.4290 ***
(0.0351) (0.0344) (0.0342)
Log(Rainfall) 0.0085 -0.00125 -0.0041
(0.00502) (0.00533) (0.0063)
Workday 0.254 *** 0.211 *** 0.169 ***
(0.0186) (0.018) (0.0201)
Month -0.000382 -0.00979 -0.0135
(0.00674) (0.00537) (0.00682)
Number of Obs. 1402 2292 1555
R² 0.6546 0.5907 0.5966
Joint significance 180.00 *** 166.52 *** 151.85 ***
The regression includes a dummy for each year from 2003 to 2007 and a constant. Driscoll-Kraay
standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01
31
Table 5: Estimation of the logarithm of Nitrogen Oxides (NOX) daily average concentration
The regression includes a dummy for each year from 2003 to 2007 and a constant. Driscoll-Kraay
standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01
Dependent Variable: (1) (2) (3)
Log(NOX) 0 - 2.5 km 2.5 - 10.0 km 10.0 - 30.0 km
Metrobus -0.1220 -0.1290 ** -0.1810 ***
(0.0545) (0.0522) (0.0449)
Temporal lag: Log(NOX) 0.4480 *** 0.3770 *** 0.4650 ***
(0.0257) (0.0199) (0.0237)
Humidity 0.0023 -0.000742 0.0036
(0.00162) (0.00103) (0.00197)
Temperature -0.0154 -0.0234 *** -0.0050
(0.00871) (0.00499) (0.0105)
Wind Direction -0.0003 -0.000083 -0.0008 **
(0.000181) (0.000124) (0.000255)
Log(Wind Speed) -0.4060 *** -0.4620 *** -0.4340 ***
(0.0339) (0.0245) (0.0322)
Log(Rainfall) -0.0014 -0.00113 -0.0071
(0.00471) (0.0045) (0.00581)
Workday 0.315 *** 0.265 *** 0.273 ***
(0.0142) (0.0146) (0.0182)
Month -0.00179 -0.00418 -0.0188 **
(0.00521) (0.00454) (0.00705)
Number of Obs. 1103 2313 1883
R² 0.6063 0.5642 0.5393
Joint significance 161.43 *** 141.73 *** 125.99 ***
32
Table 6: Estimation of the logarithm of Particulate Matter with less than 2.5 µm (PM2.5) daily
average concentration
Dependent Variable: (1) (2) (3)
Log(PM2.5) 0 - 2.5 km 2.5 - 10.0 km 10.0 - 30.0 km
Metrobus -0.2080 *** -0.2530 ** -0.3900 ***
(0.0787) (0.0855) (0.096)
Temporal lag: Log(PM2.5) 0.4320 *** 0.4400 *** 0.5130 ***
(0.0447) (0.0291) (0.0507)
Humidity -0.0176 *** -0.00854 ** -0.0023
(0.00237) (0.00195) (0.00429)
Temperature -0.0261 ** 0.0087 0.0494 **
(0.0112) (0.0092) (0.0211)
Wind Direction 0.0003 0.0003540 -0.0009
(0.000495) (0.000241) (0.000791)
Log(Wind Speed) -0.5850 *** -0.5990 *** -0.5700 ***
(0.069) (0.0548) (0.0858)
Log(Rainfall) 0.0125 0.00996 0.0415 **
(0.0108) (0.00683) (0.0161)
Workday 0.120 *** 0.1240 *** 0.223 ***
(0.0305) (0.0250) -0.043
Month 0.0167 ** 0.00136 -0.0429 ***
(0.00847) (0.00695) (0.0155)
Number of Obs. 328 1416 235
R² 0.6295 0.5678 0.6205
Joint significance 49.65 *** 70.85 *** 35.68 ***
The regression includes a dummy for each year from 2003 to 2007 and a constant. (1) & (3) use robust
standard errors and (2) uses Driscoll-Kraay standard errors. S.E. in parentheses. * p<0.10, ** p<0.05, ***
p<0.01
33
Table 7: Estimation of the logarithm of Particulate Matter with less than 10 µm (PM10) daily
average concentration
Dependent Variable: (1) (2) (3)
Log(PM10) 0 - 2.5 km 2.5 - 10.0 km 10.0 - 30.0 km
Metrobus -0.2440 * -0.1770 ** -0.1550 *
(0.0813) (0.0637) (0.0541)
Temporal lag: Log(PM10) 0.4390 *** 0.4410 *** 0.4670 ***
(0.0376) (0.0300) (0.0285)
Humidity -0.0125 ** -0.0137 *** -0.0166 ***
(0.0028) (0.00166) (0.00205)
Temperature 0.0142 0.0168 0.0180
(0.0107) (0.00871) (0.00865)
Wind Direction -0.0005 -0.0000011 -0.0007 *
(0.0003) (0.000167) (0.00028)
Log(Wind Speed) -0.3300 ** -0.2660 *** -0.2180 ***
(0.0526) (0.0310) (0.0361)
Log(Rainfall) 0.0188 0.0096 0.0091
(0.0074) (0.0059) (0.00648)
Workday 0.1400 ** 0.1920 *** 0.1420 ***
(0.0211) (0.0202) (0.0185)
Month 0.0093 0.00117 -0.000677
-0.0073 (0.00562) (0.00595)
Number of Obs. 1047 1643 1007
R² 0.5485 0.5248 0.5804
Joint significance 94.04 *** 110.22 *** 116.86 ***
The regression includes a dummy for each year from 2003 to 2007 and a constant. Driscoll-Kraay
standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01
34
Table 8: Estimation of the logarithm of Sulfur Dioxide (SO2) daily average concentration
Dependent Variable: (1) (2) (3)
Log(SO2) 0 - 2.5 km 2.5 - 10.0 km 10.0 - 30.0 km
Metrobus -0.2140 -0,1760 -0,2580
(0.177) (0.173) (0.179)
Temporal lag: Log(SO2) 0.5610 *** 0,4440 *** 0,4220 ***
(0.038) (0.026) (0.0308)
Humidity -0.0027 -0,00637 * -0,0134 **
(0.00396) (0.00348) (0.00406)
Temperature -0.0006 0,0034 0,0108
(0.0198) (0.0201) (0.0213)
Wind Direction 0.0006 0,00202 *** 0,0025 ***
(0.000639) (0.000446) (0.000643)
Log(Wind Speed) -0.3760 * -0.496 *** -0,4090 ***
(0.12) (0.0795) (0.101)
Log(Rainfall) 0.0223 -0,00631 0,0120
(0.0165) (0.013) (0.0179)
Workday 0.267 ** 0,212 *** 0,143 **
(0.0513) (0.0445) (0.0551)
Month -0.00295 -0,0153 -0,0161
(0.018) (0.0133) (0.0141)
Number of Obs. 1344 2987 1867
R² 0.3849 0,3326 0,2912
Joint significance 42.96 *** 62.42 *** 33.36 ***
The regression includes a dummy for each year from 2003 to 2007 and a constant. Driscoll-Kraay
standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01
35
Table 9: Summary of the impact of the Metrobus implementation on the different pollutants
(1) Less (2) Between (3) Between
than 2.5 km 2.5km - 10km 10km - 30km
CO DK -0.1940 ** -0.1720 *** -0.1660 **
(0.0506) (0.0394) (0.0456)
PCSE -0.2040 *** -0.2000 *** -0.1760 ***
(0.0426) (0.0460) (0.0473)
NOX DK -0.1220 -0.1290 ** -0.1810 ***
(0.0545) (0.0522) (0.0449)
PCSE -0.1610 *** -0.1440 *** -0.1590 ***
(0.0438) (0.0408) (0.0478)
PM2.5 DK -0.2080 *** -0.2530 ** -0.3900 ***
(0.0787) (0.0855) (0.096)
PCSE -0.2080 *** -0.2980 *** -0.3900 ***
(0.0756) (0.0602) -0.1020
PM10 DK -0.2440 * -0.1770 ** -0.1550 *
(0.0813) (0.0637) (0.0541)
PCSE -0.2160 *** -0.0960 * -0.1570 **
(0.0568) (0.0574) (0.0619)
SO2 DK -0.2140 -0.1760 -0.2580
(0.177) (0.173) (0.179)
PCSE -0.2770 ** -0.2310 ** -
(0.124) (0.116) -
Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01
36
Table 10: Estimated coefficients of the Metrobus implementation on the deciles of the pollutant
distributions in the area within 2.5 km around the Metrobus line 1 in Mexico-City.
Confidence Level τ (1) CO (2) NOX (3) PM2.5 (4) PM10 (5) SO2
0.90 -0.1990 *** -0.1170 * -0.3420 -0.1650 *** -0.6390 ***
(0.0443) (0.0391) (0.214) (0.0500) (0.215)
0.80 -0.1910 *** -0.1130 * -0.2620 -0.2100 *** -0.3610 ***
(0.0389) (0.0675) (0.185) (0.0412) (0.107)
0.70 -0.1830 *** -0.1040 ** -0.2350 -0.1870 *** -0.3850 ***
(0.0294) (0.0650) (0.152) (0.0486) (0.130)
0.60 -0.2080 *** -0.1100 *** -0.1730 * -0.2440 *** -0.2040
(0.0365) (0.0479) (0.0974) (0.0573) (0.128)
0.50 -0.1840 *** -0.1060 ** -0.1860 * -0.2350 *** -0.3100 **
(0.0413) (0.0447) (0.0955) (0.0504) (0.122)
0.40 -0.1830 *** -0.0941 ** -0.2270 ** -0.2560 *** -0.1710
(0.0343) (0.0385) (0.0889) (0.0567) (0.128)
0.30 -0.2190 *** -0.1020 -0.1700 -0.2090 *** -0.1880
(0.0454) (0.0418) (0.104) (0.0595) (0.166)
0.20 -0.2130 *** -0.1150 * -0.1550 ** -0.2130 *** -0.2980
(0.0443) (0.0635) (0.0716) (0.0565) (0.185)
0.10 -0.1840 *** -0.2180 *** -0.2800 *** -0.1310 * 0.0844
(0.0687) (0.0660) (0.103) (0.0713) (0.329)
Robust standard errors in parentheses. *p<0.1, **p<0.05,***p<0.01
37
Table 11: Estimated coefficients of the Metrobus implementation on the deciles of the pollutant
distributions in the area within 2.5 and 10.0 km around the Metrobus line 1 in Mexico-City.
Confidence Level τ (1) CO (2) NOX (3) PM2.5 (4) PM10 (5) SO2
0.90 -0.157 *** -0.272 *** -0.342 *** -0.139 *** -0.239 **
(0.0268) (0.0725) (0.0506) (0.0251) (0.1070)
0.80 -0.207 *** -0.18 *** -0.354 *** -0.165 *** -0.311 **
(0.0385) (0.0317) (0.0638) (0.0439) (0.1440)
0.70 -0.194 *** -0.148 *** -0.275 *** -0.102 * -0.311 ***
(0.0278) (0.0491) (0.0488) (0.0557) (0.0863)
0.60 -0.197 *** -0.125 *** -0.289 *** -0.12 *** -0.343 ***
(0.0381) (0.0370) (0.0674) (0.0411) (0.0913)
0.50 -0.209 *** -0.141 *** -0.254 *** -0.169 *** -0.338 ***
(0.0353) (0.0297) (0.0543) (0.0356) (0.0808)
0.40 -0.216 *** -0.157 *** -0.276 *** -0.195 *** -0.331 ***
(0.0439) (0.0335) (0.0574) (0.0370) (0.0720)
0.30 -0.227 *** -0.14 *** -0.252 *** -0.163 *** -0.355 ***
(0.0497) (0.0378) (0.0759) (0.0381) (0.0765)
0.20 -0.184 *** -0.131 ** -0.263 *** -0.204 *** -0.229
(0.0557) (0.0606) (0.0825) (0.0701) (0.1700)
0.10 -0.152 *** -0.0921 * -0.232 *** -0.12 -0.257 **
(0.0540) (0.0522) (0.0390) (0.0745) (0.1190)
Robust standard errors in parentheses. *p<0.1, **p<0.05,***p<0.01
38
Table 12: Estimated coefficients of the Metrobus implementation on the deciles of the pollutant
distributions in the area within 10.0 and 30.0 km around the Metrobus line 1 in Mexico-City.
Confidence Level τ (1) CO (2) NOX (3) PM2.5 (4) PM10 (5) SO2
0.90 -0.256 *** -0.187 *** -0.353 *** -0.0999 -0.147
(0.0654) (0.0546) (0.0307) (0.0983) (0.1440)
0.80 -0.171 *** -0.164 *** -0.365 *** -0.13 *** -0.313 **
(0.0564) (0.0367) (0.0610) (0.0391) (0.1390)
0.70 -0.171 *** -0.161 *** -0.281 *** -0.137 *** -0.298 ***
(0.0491) (0.0405) (0.0510) (0.0476) (0.0922)
0.60 -0.16 *** -0.166 *** -0.29 *** -0.13 ** -0.223 **
(0.0395) (0.0395) (0.0576) (0.0506) (0.0881)
0.50 -0.156 *** -0.146 *** -0.304 *** -0.151 *** -0.338 ***
(0.0427) (0.0359) (0.0488) (0.0544) (0.0783)
0.40 -0.167 *** -0.182 *** -0.281 *** -0.184 *** -0.352 ***
(0.0516) (0.0372) (0.0629) (0.0588) (0.0993)
0.30 -0.195 *** -0.191 *** -0.266 *** -0.152 ** -0.279 ***
(0.0510) (0.0411) (0.1020) (0.0622) (0.0879)
0.20 -0.145 *** -0.153 *** -0.197 ** -0.0896 -0.341 ***
(0.0532) (0.0462) (0.0818) (0.0626) (0.1200)
0.10 -0.204 *** -0.188 ** -0.319 *** -0.0892 -0.521 ***
(0.0632) (0.0773) (0.0680) (0.0692) (0.0878)
Robust standard errors in parentheses. *p<0.1, **p<0.05,***p<0.01
39
FIGURES Figure A1: Evolution of the different pollutant concentrations in the period June/2003 -
June/2007
Carbon Monoxide (CO) Nitrogen Oxides (NOX)
Particle Matter with less than 2.5 µm (PM2.5) Particle Matter with less than 10 µm (PM10)
Sulfur Dioxide (SO2)
0
0.5
1
1.5
2
2.5
3
01/0
6/20
03
01/0
9/20
03
01/1
2/20
03
01/0
3/20
04
01/0
6/20
04
01/0
9/20
04
01/1
2/20
04
01/0
3/20
05
01/0
6/20
05
01/0
9/20
05
01/1
2/20
05
01/0
3/20
06
01/0
6/20
06
01/0
9/20
06
01/1
2/20
06
01/0
3/20
07
01/0
6/20
07
Average
0 - 2.5km
Average
2.5 - 10.0km
Average
10.0 - 30.0km
0
20
40
60
80
100
120
140
160
180
01/0
6/20
03
01/0
9/20
03
01/1
2/20
03
01/0
3/20
04
01/0
6/20
04
01/0
9/20
04
01/1
2/20
04
01/0
3/20
05
01/0
6/20
05
01/0
9/20
05
01/1
2/20
05
01/0
3/20
06
01/0
6/20
06
01/0
9/20
06
01/1
2/20
06
01/0
3/20
07
01/0
6/20
07
Average
0 - 2.5km
Average
2.5 - 10.0km
Average
10.0 - 30.0km
0
10
20
30
40
50
60
70
01/0
6/20
03
01/0
9/20
03
01/1
2/20
03
01/0
3/20
04
01/0
6/20
04
01/0
9/20
04
01/1
2/20
04
01/0
3/20
05
01/0
6/20
05
01/0
9/20
05
01/1
2/20
05
01/0
3/20
06
01/0
6/20
06
01/0
9/20
06
01/1
2/20
06
01/0
3/20
07
01/0
6/20
07
Average
0 - 2.5km
Average
2.5 - 10.0km
Average
10.0 - 30.0km
0
20
40
60
80
100
120
140
01/0
6/20
03
01/0
9/20
03
01/1
2/20
03
01/0
3/20
04
01/0
6/20
04
01/0
9/20
04
01/1
2/20
04
01/0
3/20
05
01/0
6/20
05
01/0
9/20
05
01/1
2/20
05
01/0
3/20
06
01/0
6/20
06
01/0
9/20
06
01/1
2/20
06
01/0
3/20
07
01/0
6/20
07
Average
0 - 2.5km
Average
2.5 - 10.0km
Average
10.0 - 30.0km
0
5
10
15
20
25
30
35
40
01/0
6/20
03
01/0
9/20
03
01/1
2/20
03
01/0
3/20
04
01/0
6/20
04
01/0
9/20
04
01/1
2/20
04
01/0
3/20
05
01/0
6/20
05
01/0
9/20
05
01/1
2/20
05
01/0
3/20
06
01/0
6/20
06
01/0
9/20
06
01/1
2/20
06
01/0
3/20
07
01/0
6/20
07
Average
0 - 2.5km
Average
2.5 - 10.0km
Average
10.0 - 30.0km